This action might not be possible to undo. Are you sure you want to continue?
n the early 1990s, Cheyette (1992) and others introduced a separable
volatility specification of the general HeathJarrowMorton (HJM, 1992)
model. Contrary to general HJM and Libor market models, this specifi
cation allows for Markov representation of the full yield curve in a low
number of state variables. In this article, we present a class of separable
volatility structure yield curve models that incorporates stochastic volatili
ty to match the volatility smile as observed in the vanilla interest rate op
tions markets. We combine this with recent ideas for approximation of
stochastic volatility models with timedependent parameters by Piterbarg
(2005ab) to yield a fast and efficient calibration of the model.
The first sections of the article consider separable volatility structures
in HJM models and stochastic volatility models for vanilla swaptions and
caps. We then introduce our model specification and describe how cap
and swaption prices can be approximated in the model. Calibration tech
niques and numerical examples are considered. The final sections of the
article consider pricing in our model by Monte Carlo simulations and the
finite difference solution.
Separable volatility
If P(t, T) denotes the time t price of a zerocoupon bond maturing at time
T, the time t continuously compounded forward rate for deposits over the
interval [T, T + dT] is given by:
Heath, Jarrow & Morton (1992) show that any arbitragefree term structure
model with continuous evolution of the yield curve has to satisfy:
(1)
where W is a vector Brownian motion under the riskneutral measure and
{σ(t, T)}
t ≤ T
is a family of vector processes.
The HJM approach prescribes a very straightforward way of specifying
an arbitragefree term structure model that automatically fits the initial term
structure: all one needs to do is to specify the forward rate volatility struc
ture {σ(t, T)}
t ≤ T
.
However, the problem with this modelling approach is that the re
sulting model is not generally Markovian in a limited number of state vari
ables. In general, the HJM model approach requires us to use the full
forward curve as a state variable to close (1) as a Markov system. This is
independent of the dimension of the driving Brownian motion and it is
even the case if the forward rate volatility structure is deterministic. So
when we simulate the model (1) we generally need to carry forward all
points on the forward curve. Hence, the computational effort of simula
tion of the model (1) grows at a quadratic rate in the time horizon. Sim
ilarly, if we attempt to approximate the process (1) with a discrete process,
the resulting tree will be nonrecombining and thus have a number of
nodes that grow at an exponential rate in the number of time steps, or
the time horizon.
However, Cheyette (1992), Babbs (1993), Jamshidian (1991) and
Ritchken & Sankarasubramaniam (1993) independently find that if we re
strict ourselves to a volatility structure for the forward rates that are sepa
df t T t T t s ds dt t T dW t
t
T
, , , , ( ) ( )
′
( )
( )
+ ( )
′
( )
∫
σ σ σ
f t T
P t T
T
,
ln ,
( ) −
( )
∂
∂
rable in the sense that there exist a deterministic vector function g on R
k
and a matrix process h on R
k × k
, so that:
(2)
then a Markov representation of the dynamics of the yield curve, involv
ing k + k × (k + 1)/2 state variables, emerges.
Without loss of generality, the model can in this case be formulated as:
(3)
In the context of the separable formulation (2) we have:
The first k state variables, the elements of X, can be interpreted as yield
curve factors that pertubate the forward curve and are directly associated
with the driving Brownian motions, whereas the remaining k × (k + 1)/2
state variables, the elements of the symmetric matrix Y, can be seen as ‘con
vexity’ terms that have to be carried along to keep the model arbitrage free.
For k = 1 and η deterministic, Y becomes deterministic and we obtain
the general Gaussian model, that is, a Vasicek (1977) model with timede
pendent parameters. This led Jamshidian (1991) and Babbs (1993) to de
note the separable volatility specification as, respectively, ‘quasi’ and
‘pseudo’ Gaussian models.
The potential computational saving in using this type of model rather
than the general HJM approach is considerable. For example, if we con
sider the case of pricing a 30year structure with quarterly fixings and pay
ments by simulation, the general HJM or Libor market model (LMM)
approach will require the evolution of at least 120 points on the yield curve,
whereas a fourfactor version of the separable model requires the evolu
tion of a maximum of 14 state variables.
For the onedimensional case, k = 1, a finite difference solution is vi
able and is most often a more efficient numerical solution method than
Monte Carlo simulation. We will discuss this later. Finite difference solu
tion of simpler versions of the model are also considered in Andreasen
(2000) and Andersen & Andreasen (2002).
It is worth noting that if we let κ
1
, ... , κ
k
be constants, then:
for k → ∞ and an appropriately chosen sequence κ
1
, κ
2
, .... So the model
(3) can be seen as a representation of the forward rate volatility structure
σ t t e t e t d
i
i
i
k
, ( ) + ( )
′
→ ( )
−
−
∑ ∫
τ η η κ
κ τ κτ
κ
1
g T g T h t I t
g t
( ) ( ) ( ) ( )
( )
−
0
1
, , η
dX t Y t I X t dt t dW t X
dY t t t
t
( ) ( ) − ( )
( )
+ ( ) ( ) ( )
( ) ( ) ( )
′
−
( )
ι η
η η
κ
, 0 0
II Y t Y t I dt Y
P t T
P T
P t
e
t t
G t
κ κ ( ) ( )
−
( ) − ( )

.
`
}
( )
( )
( )
( )
,
,
,
,
0 0
0
0
,, , ,
, , , ,
T X t G t T Y t G t T
t
T
G t T g t s ds g t T e
( )
′
( )− ( )
′
( ) ( )
−
( ) ( ) ( )
∫
1
2
κκ κ
ι κ
1
1 1
u du u du
t
T
k
t
T
e
X G g W
( ) − ( )
∫ ∫

.
`
}
′
( )
′
∈
,...,
,..., , , , , , »»
» … »
k
k k
a k
k k
Y I a a
,
, , diag , , η∈ ( ) ∈
× ×
1
σ t T g T h t , ( )
′
( )
′
( )
104 RISK SEPTEMBER 2005
●
WWW.RISK.NET
Cutting edge
l
Interest rates
Back to the future
Current developments in exotic interest rate products push the demand for more sophisticated
interest rate models. Here, Jesper Andreasen presents a new class of stochastic volatility multi
factor yield curve models enabling quick calibration and efficient Monte Carlo simulation
In our experience, the implied skew and smile parameters m and ε are
quite stable over time, so in practice only the volatility level parameter λ
needs to be updated on a regular basis, say daily or weekly. To illustrate
this, figure 1 shows the deviations in terms of implied Black volatility from
endofmonth Totem
1
consensus quotes for euro swaptions for strikes rang
ing from 5–95% delta, for two different models (4). The first model had its
m and ε grids fitted the month before whereas the second was fitted on
the particular date. In both cases we set the λ so that the model fits the at
themoney Totem levels. Expiries range from six months to 20 years and
tenors from one year to 30 years. In total, 912 swaptions were priced. We
see that both models for the most part agree with the Totem consensus
quotes within +/–0.25% in BlackScholes implied volatility for all strikes.
Figure 2 gives a further indication of the stability of the skew and smile
parameters of the model by showing the evolution of monthly calibrated
on a (discrete) basis of exponential functions. The function κ → η
κ
(t) can
thus be viewed as the inverse Laplace transform of the forward rate volatil
ity structure in the tenor dimension: t → σ(t, t + τ).
Stochastic volatility processes
The most popular stochastic volatility model for caps and swaptions appears
to be the SABR model by Hagan et al (2002) where the volatility is specified
as a geometric Brownian motion that has some correlation with the under
lying forward swap rate. This model is quite difficult to work with in the
context of full yield curve models, for a number of reasons.
First, the SABR model does not incorporate meanreversion in volatili
ty, which means that when the model is fitted to observed cap and swap
tion prices the implied volatility of volatility parameter most often turns
out to be decreasing with the expiry of the underlying option. This in turn
implies that a full dynamic version of the SABR model would have to ex
hibit even steeper decreasing forward volatility of volatility. Second, in
many implementations of the SABR model the correlation between volatil
ity and underlying rate are quite different for different expiries and tenors.
Nonzero correlation is technically quite difficult to handle in a full yield
curve model and potentially timevarying correlation is of course even
more complicated. Third, as the SABR model has no closedform for Eu
ropeanstyle option prices, it is typically implemented for European op
tion pricing by expansion techniques whose accuracy deteriorates for
longer expiries. This may have limited practical importance if the SABR
model is only used for Europeanstyle option pricing, but our scope is to
price general pathdependent instruments, so we need our Europeanstyle
option pricing to be consistent with the actual specified dynamics.
Instead we follow Andersen & Andreasen (2002) and use the following
model as our basis for developing a full yield curve model with stochas
tic volatility:
(4)
where W
A
is a Brownian motion under annuity measure, that is, the mar
tingale measure with the annuity:
δ
i
= t
i
– t
i – 1
is the day count fraction, S(t) = (P(t, t
0
) – P(t, t
n
))/A(t) is the for
ward par swap rate under consideration, and all the parameters λ, m, ε, β
are constants. The swap rate is a martingale under the annuity measure.
In terms of the implied BlackScholes volatility smile, the level is con
trolled by λ. As correlation between the swap rate and the volatility is as
sumed to be zero, the slope of the smile is fully controlled by the m
parameter. The smile becomes increasingly negatively sloped as m is de
creased. Subnormal skews, corresponding to m < 0, are possible with the
note of caution that S is restricted from above by
1 – m
m
S(0) when m is neg
ative. Increasing the volatility of local variance, ε, increases the curvature
of the smile. Increasing the speed of meanreversion, β, increases the rate
at which the curvature of the smile decays with expiry.
The model is essentially a ‘shifted’ Heston (1993) model, so it allows
for an analytic solution based on numerical inversion of the Fourier trans
form. Lipton (2002) and Lewis (2000) give representations of the option
price that avoid the numerical instability of the representation in the orig
inal Heston (1993) paper.
This model gives a good fit to observed cap and swaption prices with
reasonably stable parameters across expiries and tenors. An example of
the fitted m, ε parameters is given in table A. We note that the model’s
volatility of variance parameter, ε, is related to lognormal volatility of volatil
ity by the approximate relation:
lognormal volatility of volatility ≈ ε / 2
A t P t t
i
i
n
i
( ) ( )
∑
δ
1
,
dS t z t mS t m S dW t
dz t z t dt z t
A
( ) ( ) ( ) + − ( ) ( )
]
]
( )
( ) − ( ) ( )
+
λ
β ε
1 0
1
1
(( ) ( )
dW t
dW dW
A
A A
2
1 2
0
WWW.RISK.NET
●
SEPTEMBER 2005 RISK 105
m 6m 1y 2y 5y 10y 15y 20y 30y
6m 0.75 0.75 0.62 0.50 0.44 0.38 0.34 0.29
1y 0.75 0.65 0.55 0.46 0.38 0.32 0.30 0.27
3y 0.68 0.58 0.49 0.37 0.31 0.26 0.25 0.22
5y 0.65 0.52 0.43 0.32 0.27 0.23 0.21 0.17
10y 0.58 0.45 0.38 0.27 0.23 0.21 0.17 0.14
15y 0.48 0.36 0.31 0.26 0.20 0.18 0.15 0.13
ε 6m 1y 2y 5y 10y 15y 20y 30y
6m 1.15 1.13 1.13 1.18 1.29 1.29 1.30 1.27
1y 1.15 1.01 1.05 1.06 1.11 1.14 1.13 1.12
3y 1.05 0.93 0.91 0.93 0.94 0.93 0.94 0.93
5y 0.95 0.88 0.88 0.86 0.85 0.86 0.86 0.85
10y 0.86 0.89 0.89 0.87 0.84 0.84 0.84 0.83
15y 1.03 1.00 0.97 0.94 0.91 0.90 0.91 0.89
Note: this table reports best fit m and ε parameters to observed cap and
swaption prices for β = 0.05. Expiries are in the rows and tenors are in the
columns. The currency is the euro and the parameters were estimated
from Totem consensus prices for the end of a particular month in 2004
A. Skew and smile parameters fitted to euro
cap and swaption prices
1. This and last month’s models against
Totem quotes
–1.0
–0.8
–0.6
–0.4
–0.2
0.0
0.2
0.4
0.6
0.8
1.0
0% 20% 40% 60% 80% 100%
mdl(t – 1m)
mdl(t)
%
This figure shows the deviations from Totem consensus quotes for
euro swaptions in terms of implied Black volatility for two models. One
that had its m, ε parameters fitted the month before the other had its
parameters fitted this month. Expiries range from six months to 20
years, tenors from one year to 30 years, and strikes from 5% to 95%
in BlackScholes delta terms – all in all, 912 swaptions. All data is as
of a particular end of month in 2004
1
Totem provides independent mark verification services for interbank options based on
midmarket quotes from approximately 20 leading option dealers
m and ε for the 10year × 10year euro swaption over 2003–05. Though
the implied skew parameter mincreases over the period, the implied volatil
ity of variance ε is more or less constant over the period and both time se
ries exhibit low volatility.
Model specification
Andersen & Andreasen (2002) suggest a Libor market model with stochastic
volatility, which is extended by Piterbarg (2003) to allow for a time and
tenordependent local volatility skew parameter. The motivation for this is
that if we consider implied parameters of the model (4), as in table A, we
typically see that the skew parameter m is fairly constant across expiry but
it tends to decrease with tenor. On the other hand, the implied ε parame
ter appears to be fairly constant across both expiry as well as tenor, at least
for expiries over one year.
If we use continuously compounded rates rather than discrete rates as
model primitives, the Piterbarg model can be formulated as:
(5)
where m, λ and ρ are deterministic functions of time and maturity, ε is
a deterministic function of time and β is a constant. We note that m = 0
corresponds to a normal model whereas m = 1 corresponds to a log
normal model.
Fix k tenors τ
1
, ... , τ
k
. For the corresponding forward rates we have:
(6)
dF t z t I I I I I I R t dW t O d
M t F t M t F t
( ) ( ) + −
( )
]
]
]
( ) ( ) +
( ) ( ) ( ) ( ) ( ) 0 λ
tt
F t f t t f t t
M t m t t m t t
n
k
( )
( ) + ( ) + ( ) ( )
′
∈
( ) + ( ) +
, , , ,
, , , ,
τ τ
τ
1
1
… »
… ττ
λ τ λ τ
ρ
λ
n
k
t n
k k
I t t t t
R t t t
( ) ( )
′
∈
+ ( ) + ( ) ( )
∈
( )
( )
×
»
… » Diag , , , ,
,
1
++ ( )
′
+ ( )
′

.
`
}
∈
×
τ ρ τ
1 1
; ; , … » t t
k k
df t T z t T m t T f t m t T f T
t T t T
, , , , ,
, ,
( ) ( ) ( ) ( ) + − ( ) ( ) ( )
]
]
× ( ) ( )
1 0
λ ρ
′′
( ) + ( )
( ) − − ( ) ( ) ( )
+ ( ) ( ) ( )
( ) ∈
dW t O dt
dz t z z t dt z t t dZ t
t T
β ε
λ
1 1
, , » ρρ ρ t T t T
Z t dZ t dW t
k
, , ,
,
( ) ∈ ( )
( ) ∈ ( ) ( )
»
»
1
0
Here the matrix product RR′ is the instantaneous correlation matrix for the
k forward rates.
Under the separable volatility specification in (3), we have:
(7)
Equating diffusion terms of (6) and (7) yields:
(7a)
with:
(7b)
The volatility specification (7) in combination with (3) defines our model.
In most cases we choose constant κ
1
, ... , κ
k
as well as a constant cor
relation structure RR′. The latter is typically estimated from the historical
timeseries data of the yield curve. In this case, the model parameters that
need to be set by calibration to swaption and cap prices are:
■ The forward rate volatility structure, λ for all times t and the tenors
τ
1
, ... , τ
k
.
■ The forward rate skew structure, m for all times t and the tenors
τ
1
, ... , τ
k
.
■ The forward volatility of volatility, ε for all times t.
In terms of the implied BlackScholes volatility smiles for swaptions and
caplets, the first parameter controls the absolute level, the second the slope
(skew) and the third the curvature (smile).
We see that the model, at least in principle, can exactly fit the volatili
ty level and slope for all expiries along k tenors, whereas the curvature
can only be fitted exactly for one tenor. In practice, though, our calibra
tion will most often be on a best fit basis.
For the onefactor case, k = 1, we do, however, often choose to go for
an exact fit to a specific strip of swaptions or caplets. In this case we often
specify the model a bit differently, namely:
(8)
where λ, m are now scalar functions of time and S is a par swap rate re
ferring to different swap periods over the time horizon. For example, if we
choose to fit the model to the strip of 1×29, 2×28, … , 29×1 swaption
smiles, we let S be the 1×29 par swap rate for times between year zero
and one, 2×28 par swap rate for times between year one and two, up to
29×1 par swap rate for times between year 28 and 29.
Swaption pricing
For efficient calibration of the model, closedform pricing of caps and swap
tions is essential. In this section, we describe an accurate (near) closed
form approximation.
Using Itô’s lemma and the fact that the swap rate S is a martingale under
the annuity measure, we get:
(9)
where we let subscripts denote partial derivatives, that is, S
X
= (∂S/∂X
1
, ...
, ∂S/∂X
k
). Given fixed meanreversion coefficients κ
1
, ... , κ
k
this derivative
can be calculated in closed form by combining (7) with the bond price
formula in (3).
Our approximation goes in two steps:
■ A. Approximate the stochastic differential equation (9) by the model:
dS t S t t dW t
X
A
( ) ( )
′
( ) ( ) η
η λ t z t m t S t m t S t ( ) ( ) ( ) ( ) + − ( ) ( ) ( )
]
]
( ) 1 0
dz t z t dt t z t dZ t dZdW ( ) − ( ) ( )
+ ( ) ( ) ( ) β ε 1 0 ,
Γ t t z t I I I I I I R t
t
M t F t M t F t
( ) ( ) ( ) + −
( )
]
]
]
( )
⇓
( )
( ) ( ) ( ) ( ) ( )
η
η
λ 0
( ) ( ) + −
( )
]
]
]
( )
−
( ) ( ) ( ) ( ) ( )
z t t I I I I I I R t
M t F t M t F t
Γ
1
0 λ
dF t t t dW t O dt
t
g t t
g t t
k
( ) ( ) ( ) ( ) +
( )
+ ( )
′
+ ( )
′
]
]
Γ
Γ
η
τ
τ
( )
,
,
]]
]
]
]
∈
×
»
k k
106 RISK SEPTEMBER 2005
●
WWW.RISK.NET
Cutting edge
l
Interest rates
2. Timeseries evolution of calibrated m, ε
parameters
–0.2
0
0.2
0.4
0.6
0.8
1.0
1 3 5 7 9 11 13 15 17 19 21 23 25
m
eps
L
e
g
e
n
d
?
This figure shows the evolution of the monthly calibrated 10year by
10year skew and smile parameters m, ε over the period 2003–05.
The currency is the euro
here for space considerations, but the main point is that the technique is
both very quick and accurate. Computationally, the method relies on a nu
merical solution of one Ricatti ordinary differential equation per swaption
pricing  all remaining calculations are done in closed form. Compared
with a direct solution of (10) by numerical inversion of Fourier transform
as suggested in Andersen & Andreasen (2002), this technique is much faster
and only marginally less accurate.
Calibration
We start by fixing κ
1
, ... , κ
k
, the correlation structure for the forward rates
RR′, and a set of tenors τ
1
, ... , τ
k
of the model. We further fix a time grid
0 = t
0
< t
1
< ... of expiries and a set of tenors {T
j
} corresponding to the
swaption smiles that we wish to calibrate the model to. We assume that
we have fitted parameters λ
~
hj
, m
~
hj
, ε
~
hj
of the model (4) for these expiries
(h) and tenors (j) of the calibration swaptions, as in table A.
We let the model (3) and (7) be parameterised by:
for t
h – 1
< t ≤ t
h
. We use approximation A and B to give us constant para
meters λ
~
hj
, m
~
hj
, ε
~
hj
for each swaption. We now calibrate the model by boot
strapping, that is, we solve the optimisation problems:
sequentially for h = 1, 2, .... Here γ
l
, γ
m
, γ
ε
are weights for balancing the dif
ferent objectives against each other. Most often we calibrate the model in
a sequence where only one of the weights γ
l
, γ
m
, γ
ε
is nonzero at the time.
As an example of this, consider simultaneous calibration of a fourfac
tor model of the type specified in (3) and (7) to all the euro cap and swap
tion implied volatility smiles of 19 expiries ranging from six months to 20
years and eight tenors ranging from six months to 30 years. The implied
volatility smiles are parameterised by the parameters in table A. We set:
and use a correlation matrix estimated for historical timeseries data of for
ward rate curves.
The resulting model parameters are shown in table B. We see that the
forward skew parameters, m
i
, are decreasing more sharply in tenor than
the corresponding ‘term’ skew parameters shown in table A. This is con
sistent with the findings in Piterbarg (2003). There does not appear to be
a clear trend over time in any of the calibrated parameters. However, there
is more noise in the calibrated forward skew parameters than in the Piter
barg (2003) case. This is probably due to the fact that we make no attempts
to smooth our calibrated parameters in the time dimension. The calibra
tion takes about five seconds of computer time.
κ κ κ κ
τ τ τ τ
1 2 3 4
1 2 3 4
0 015 0 15 0 30 1 20
6 2 10
, , , . , . , . , .
, , , , ,
( ) ( )
( ) m y y y y ,30 ( )
min
{ , , }
, ,
λ ε
λ ε
γ λ λ γ γ
hi hi h i k
m
hj
hj
j
m
hj
hj
j
m m
−
( )
+ −
( )
+
∑ ∑
1
2 2
…
εε ε
hj
hj
j
−
( ) ∑
2
λ τ λ τ ε ε t t m t t m t
i hi i hi h
, , , , + ( ) + ( ) ( )
(10)
where all parameters are timedependent.
■ B. Approximate the stochastic differential eqaution (10) by the timeho
mogeneous model:
(11)
where all parameters are constant.
Approximation A essentially involves finding timedependent parame
ters λ
_
, m
_
so that the diffusion in (9) is approximated by the diffusion in
(10), that is:
(12)
Equating levels in (12) at X(t) = Y(t) = 0 yields:
(13)
Differentiating (12) with respect to X
i
at X(t) = Y(t) = 0 yields:
(14)
for i = 1, ... , k. Due to the form of η, the righthand side of (14) is inde
pendent of z(t), so (14) forms k linear equations in m
_
. We solve these by
regression:
(15)
All quantities in (13) and (15) can be calculated in closed form using the
zerocoupon bond price formula in (3).
It should be noted that this approximation can be slightly refined by
evaluating (13) and (15) along levels of X, Y corresponding to approxi
mate expected levels of X, Y under the annuity measure of the swaption
under consideration.
Approximation B involves finding constant parameters so that the model
(11) produces option prices that are close to those of (10) with parame
ters given by (13) and (15). We use the methodology suggested by Piter
barg (2005ab). The exact details are quite complicated and are omitted
m t
S t
X
z t S t t
t S
X
i
X
X Y
i
k
i
( )
( )
∂
∂
( ) ( )
′
( )
]
]
]
]
( )
−
∑
1
2
0 0
1
2
η
λ
,
00
2
0 0
1
( ) ( )
]
]
]
∑
S t
X
X Y
i
k
i
,
λ η t S S t m t
z t X
S t t
X
X Y
i
X
i
( ) ( ) ( )
]
]
( )
( )
( )
∂
∂
( )
′
( )
]
]
]
2
0
2
0
1
]]
X Y 0 0 ,
λ
λ
t
S
S t t I I R
X F
X Y
( )
( )
( )
′
( )
−
( )
2
2
1
0
0 0
2
1
0
Γ
,
z mS m S S
X
λ η
2
2
2
1 0 + − ( ) ( )
]
]
≈ ′
dS t z t mS t m t S dW t
d z t z
A
( ) ( ) ( ) + − ( )
( )
( )
]
]
]
( )
( ) −
λ
β
1 0
1
1
t dt z t dW t
A
( )
( )
+ ( ) ( ) ε
2
dS z t m S m S dW
dz z
t t t t t t
t t
A
( ) ( ) ( ) ( ) ( ) ( ) ( )
]
]
( )
( ) (
+ −
−
λ
β
( ) 1
1
0
1
)) ( ) ( ) ( ) ( ) + dt z dW t t t
A
ε
2
WWW.RISK.NET
●
SEPTEMBER 2005 RISK 107
Risk welcomes the submission of technical articles on topics relevant to our
readership. Core areas include market and credit risk measurement and man
agement, the pricing and hedging of derivatives and/or structured securities, and
the theoretical modelling and empirical observation of markets and portfolios.
This list is not an exhaustive one.
The most important publication criteria are originality, exclusivity and rele
vance – we attempt to strike a balance between these. Given that Risk techni
cal articles are shorter than those in dedicated academic journals, clarity of
exposition is another yardstick for publication. Once received by the technical
editor and his team, submissions are logged, and checked against the criteria
above. Articles that fail to meet the criteria are rejected at this stage.
Articles are then sent to one or more anonymous referees for peer review. Our
referees are drawn from the research groups, risk management departments and
trading desks of major financial institutions, in addition to academia. Many have
already published articles in Risk. Depending on the feedback from referees, the
technical editor makes a decision to reject or accept the submitted article. His deci
sion is final.
We also welcome the submission of brief communications. These are also
peerreviewed contributions to Risk but the process is less formal than for full
length technical articles. Typically, brief communications address an extension
or implementation issue arising from a fulllength article that, while satisfying
our originality, exclusivity and relevance requirements, does not deserve full
length treatment.
Submissions should be sent to the technical team at technical@
incisivemedia.com. The preferred format is MS Word, although Adobe PDFs are
acceptable. The maximum recommended length for articles is 3,500 words, and
for brief communications 1,000 words, with some allowance for charts and/or for
mulas. We expect all articles and communications to contain references to previ
ous literature. We reserve the right to cut accepted articles to satisfy production
considerations. Authors should allow four to eight weeks for the refereeing process.
Guidelines for the submission of technical articles
The error of such a calibration can be split in two. First, there is the error
from the fact that a fourfactor model will not be able to exactly match the
smiles of eight tenors. We show this error by pricing swaptions and caps
for all the calibration expiries and tenors by use of the approximation A
and B, and comparing the resulting prices to those of the target model. The
strikes chosen correspond to 5–95% delta in BlackScholes terms. In all, we
price 1,024 caplets and swaptions. We call this error ‘pure calibration error’
and it is shown in figure 3. We see that the pure calibration error is within
+/–0.25% in BlackScholes volatility terms in most of the range.
What actually counts, however, is of course what the error is when the
model is simulated. We call this “total calibration error” and the result of
pricing up all the calibration swaptions by simulation is shown in Figure
4. We see that the total calibration error is within +/0.40% in BlackSc
holes volatility terms in most of the range.
In summary, a fourfactor version of the model can simultaneously fit
market prices of caps and swaptions for all strikes (5–95% delta), expiries
(six months to 20 years), and tenors (six months to 30 years), within a tol
erance of 0.4% in implied Black volatility terms. Moreover, the calibration
only takes about five seconds of computer time.
Monte Carlo simulation
Strictly speaking, stochastic differential equations of the type defined by
(3) and (7) can in some cases exhibit explosive behaviour. To avoid this
problem, we follow Heath, Jarrow & Morton (1992) and simply replace f(t,
t + τ
i
) in (7) with:
(16)
where c is some constant.
Due to the fact that the natural domain for the stochastic volatility fac
tor z is {z ≥ 0}, straightforward Euler discretisation of the stochastic dif
ferential equation for z is going to exhibit very poor convergence as we
decrease the time steps ∆t → 0. Instead, we prefer to use the following
(local) lognormal discretisation:
f t t f t c f t t f t c
i i i i
, max , , min , , , + ( ) + ( ) − + ( ) + ( ) +
( ) ( )
τ τ τ τ 0 0
108 RISK SEPTEMBER 2005
●
WWW.RISK.NET
Cutting edge
l
Interest rates
t λ
1
(t) λ
2
(t) λ
3
(t) λ
4
(t) m
1
(t) m
2
(t) m
3
(t) m
4
(t) ε(t)
0.5 0.2120 0.2452 0.1063 0.0907 0.74 0.47 –0.28 –0.44 1.22
1 0.2242 0.2198 0.1262 0.1017 0.82 0.32 –0.05 –0.56 0.97
2 0.2266 0.2038 0.1325 0.0992 0.92 0.33 –0.01 –0.74 0.95
3 0.2464 0.1869 0.1336 0.0873 0.89 0.30 0.01 –1.01 0.80
4 0.2602 0.1906 0.1276 0.1021 0.89 0.26 0.01 –0.84 0.82
5 0.2693 0.1664 0.1278 0.0784 0.89 0.22 0.05 –1.65 0.75
6 0.2954 0.1678 0.1246 0.0872 0.83 0.24 0.06 –1.10 0.86
7 0.2798 0.1490 0.1268 0.0805 0.87 0.26 0.08 –1.07 0.85
8 0.3064 0.1496 0.1185 0.0635 0.80 0.22 0.10 –1.48 0.85
9 0.3078 0.1335 0.1207 0.0482 0.80 0.25 0.11 –1.84 0.84
10 0.3040 0.1183 0.1172 0.0410 0.81 0.25 0.13 –1.88 0.84
11 0.2934 0.1162 0.1278 0.0416 0.82 0.26 0.13 –2.01 0.97
12 0.2719 0.1148 0.1219 0.0602 0.87 0.26 0.16 –0.99 0.99
13 0.2541 0.0917 0.1276 0.0519 0.90 0.35 0.16 –1.01 1.02
14 0.1891 0.0798 0.1236 0.0658 1.40 0.37 0.20 –0.50 1.04
15 0.1574 0.0529 0.1254 0.0707 1.83 0.70 0.20 –0.36 1.06
16 0.1671 0.1009 0.1410 0.0695 1.73 0.19 0.20 –1.42 0.94
17 0.1705 0.1013 0.1413 0.0734 1.77 0.35 0.14 –0.89 0.94
18 0.1690 0.0893 0.1492 0.0623 1.97 0.45 0.12 –1.01 0.94
20 0.1147 0.0768 0.1554 0.0631 2.03 1.70 –0.11 0.39 0.94
Note: this table reports the resulting parameters when calibrating a four
factor model to the euro swaption and cap data of table A
B. Parameters of a calibrated four
factor model
Maturity LMM HJM
5y 2.12 1.14
10y 7.20 2.22
15y 15.19 3.33
20y 26.21 4.46
25y 40.27 5.53
30y 55.13 6.56
Note: CPU times in seconds for simulation of 5y, … , 30y vanilla interest
rate swaps with monthly reset in a fourfactor Libor market model and our
fourfactor separable volatility structure HJM model
C. CPU times for simulation in LMM and in
separable HJM
3. Pure calibration error
–1.0
–0.8
–0.6
–0.4
–0.2
0.0
0.2
0.4
0.6
0.8
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
%
This figure shows the pure calibration error in terms of implied Black
Scholes volatility when calibrating a fourfactor model to the full euro
cap and swaption market. We depict the difference between the yield
curve model and the target when we price caps and swaptions under
our approximations A and B. Expiries range from six months to 20
years, tenors from six months to 30 years and strikes from 5–95% in
terms of BlackScholes delta – all in all 1,024 caplets and swaptions
4. Total calibration error
–0.8
–0.6
–0.4
–0.2
0.0
0.2
0.4
0.6
0.8
1.0
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
%
This figure shows the total calibration error in terms of the Black
Scholes volatility when calibrating a fourfactor model to the full euro
market. We depict the difference between the yield curve model and
the target when we price caps and swaptions by simulation. 131,072
simulations were used, making the simulation error roughly of the
order of 0.10% in terms of implied BlackScholes volatility. Expiries
range from six months to 20 years, tenors range from six months to
30 years and strikes range from 5–95% BlackScholes delta – all in
all 1,024 caplets and swaptions
but this does not seem to be a problem in practice.
In summary, we have a scheme with the following properties:
■ Uniform von Neuman stability.
■ Accuracy of O(∆t
2
+ ∆x
2
+ ∆y
4
+ ∆z
p
), p < 2.
■ Workload of O(∆t
–1
× ∆x
–1
× ∆y
–q
× ∆z
–1
), q > 1.
In practice, a 30year Bermuda swaption is accurately priced on a grid of
dimension 50 × 100 × 10 × 15 (t × x × y × z) steps and this takes about
three seconds of computer time.
Conclusion
We have presented a class of stochastic volatility yield curve models with
quick and accurate calibration and significantly quicker Monte Carlo sim
ulation than general HJM or Libor market models. A onefactor version of
the model can be implemented with a finite difference solution and can
thus be used as an alternative to the standard onefactor models for day
today management of large portfolios of interest rate exotics. ■
Jesper Andreasen is a principal in the fixed income quantitative
research group at Bank of America in London. Email:
jesper.andreasen@bankofamerica.com
where we choose z
_
, v so that the lognormal approximation matches the
two first conditional moments of z(t
h + 1
) given z(t
h
), that is:
We combine this with standard Euler discretisation of X, Y. With typical
parameter values, accurate pricing can be obtained with monthly or quar
terly time stepping.
The strength of the separable volatility structure relative to the general
HJM or LMM specification is the speed in simulation of the model. To il
lustrate this, we perform simulation of vanilla swaps with monthly rate
reset in two models: an LMM with four factors and our separable model
also with four factors. The resulting computer times are reported in table
C. We see that in the LMM the computational time increases roughly with
the square of the simulation horizon whereas it is linear for the separable
model. Table C and our experience indicate that one can obtain compu
tational savings of up to a factor 10 for longerdated structures with the
separable model relative to the LMM.
Finite difference solution
For the onefactor model, k = 1, finite difference solution is an efficient al
ternative to Monte Carlo simulation. The associated pricing partial differ
ential equation can be written as:
We use an alternating direction implicit scheme (see Mitchell & Grif
fiths, 1980) that splits the solution over each time step into three steps:
(17)
where V(t) is to be interpreted as a threedimensional tensor of values at
time t.
We use the standard threepoint discretisation for D
x
and D
z
, but for
D
y
we use a fivepoint discretisation for the first derivative. This gives high
er accuracy in the y dimension, O(∆y
4
), and enables us to get away with
relatively few y steps, say 10. The disadvantage of the fivepoint discreti
sation is that the workload increases at a rate higher than the O(∆y
–1
) of
a threepoint scheme but we find that is worth it in this particular case.
Squareroot processes such as (7b), with high volatility and low mean
reversion and therefore high probability of hitting z = 0 can be tricky to
solve numerically. Linear discretisation of the z axis according to the stan
dard deviation of z at maturity leads to very few points in the interval [0,
1] relative to the number of points between one and the upper bound of
z. Attempting to solve this problem by transforming the state variable in
troduces infinite drift for the transformed variable at z = 0 and this is there
fore not a recommendable route. Instead we choose to discretise z according
to z
j
O(j
2
). This means that we get lower asymptotic accuracy than O(∆z
2
)
1 1
2
2
3
1 1
2
1
∆
∆
∆
∆
∆
t
D V t t
t
D D D V t t
t
x x y z
−
]
]
]
+

.
`
}
+ + +
]
]
]
+ ( )
−−
]
]
]
+

.
`
}
+

.
`
}
− + ( )
−
1
2
1
3
1 2
3
1
2
1
D V t t
t
V t t D V t t
t
y y
∆
∆
∆ ∆
∆
11
2
1 1
3
1
2
D V t
t
V t t D V t t
z z
]
]
]
( ) +

.
`
}
− + ( )
∆
∆ ∆
0
3
1
2
3
2
2
2
2
2
∂
∂
+ + +
]
]
− + − + ( )
∂
∂
+
∂
∂
− + −
V
t
D D D V
D
r
x y
x x
D
r
x y z
x
y
κ η
η κκ
β ε
y
y
D
r
z
z
z
z
z
( )
∂
∂
− + − ( )
∂
∂
+
∂
∂ 3
1
1
2
2
2
2
z e z t
v z e z t e
t
t t
+ ( ) −
( )
+ −
( )
+ ( ) −
( )
−
− ∆
− − ∆ − ∆
1 1
1
2
1 1
2 2
2 2
β
β β
ε
β
ε
β
ln ee
t − ∆
( )
¹
´
¹
¹
¹
¹
`
¹
¹
¹
]
]
]
]
2β
z t t ze
v vN
+ ∆ ( )
− + ( )
1
2
2
0 1 ,
WWW.RISK.NET
●
SEPTEMBER 2005 RISK 109
Andersen L and J Andreasen, 2002
Volatile volatilities
Risk December, pages 163–168
Andreasen J, 2000
Turbocharging the Cheyette model
Working paper, Gen Re Securities
Babbs S, 1993
Generalised Vasicek models of the term structure
Applied Stochastic Models and Data Analysis 1, pages 49–62
Cheyette O, 1992
Markov representation of the HeathJarrowMorton model
Working paper, Barra
Hagan P, D Kumar, A Lesniewski and D Woodward, 2002
Managing smile risk
Wilmott Magazine, July, pages 84–108
Heath D, R Jarrow and A Morton, 1992
Bond pricing and the term structure of interest rates: a new methodology for
contingent claims valuation
Econometrica 60, pages 77–106
Heston S, 1993
A closedform solution for options with stochastic volatility with applications to
bond and currency options
Review of Financial Studies 6, pages 327–344
Jamshidian F, 1991
Bond and option evaluation in the Gaussian interest rate model
Research in Finance 9, pages 131–170
Lewis A, 2000
Option valuation under stochastic volatility
Finance Press
Lipton A, 2002
The vol smile problem
Risk February, pages 61–65
Mitchell A and D Griffiths, 1980
The finite difference method in partial differential equations
John Wiley, New York
Piterbarg V, 2003
A stochastic volatility forward Libor model with a term structure of volatility smiles
Working paper, Bank of America, London, available from www.ssrn.com
Piterbarg V, 2005a
Smiling hybrids
Forthcoming in Risk
Piterbarg V, 2005b
Stochastic volatility model with timedependent skew
Forthcoming in Applied Mathematical Finance
Ritchken P and L Sankarasubramaniam, 1993
On finite state Markovian representations of the term structure
Working paper, Department of Finance, University of Southern California
Vasicek O, 1977
An equilibrium characterization of the term structure
Journal of Financial Economics 5, pages 177–188
REFERENCES
First.86 0.84 0.87 0.85 0.94 0.89 1.91 0.93 0.31 2y 1.36 1y 1. the slope of the smile is fully controlled by the m parameter.75 0. say daily or weekly. it is typically implemented for European option pricing by expansion techniques whose accuracy deteriorates for longer expiries.8 –1.46 0.93 0. ε. To illustrate this.68 0. This and last month’s models against Totem quotes 1. Expiries are in the rows and tenors are in the columns. as the SABR model has no closedform for Europeanstyle option prices. the implied skew and smile parameters m and ε are quite stable over time.03 1y 0.27 0.25 0.06 0.4 –0.75 0. The swap rate is a martingale under the annuity measure.86 0.21 0.93 0. Skew and smile parameters fitted to euro cap and swaption prices m 6m 1y 3y 5y 10y 15y ε 6m 1y 3y 5y 10y 15y 6m 0.44 0. An example of the fitted m.29 0. The smile becomes increasingly negatively sloped as m is decreased.15 1. Figure 2 gives a further indication of the stability of the skew and smile parameters of the model by showing the evolution of monthly calibrated 1 Totem provides independent mark verification services for interbank options based on midmarket quotes from approximately 20 leading option dealers WWW.25% in BlackScholes implied volatility for all strikes. One that had its m. so we need our Europeanstyle option pricing to be consistent with the actual specified dynamics. t + τ).29 1.00 2y 0. Increasing the volatility of local variance. tenors from one year to 30 years.2 –0.27 0. Third.30 1.15 1. m. the level is controlled by λ.18 1. increases the curvature of the smile.13 1. In both cases we set the λ so that the model fits the atthemoney Totem levels. This may have limited practical importance if the SABR model is only used for Europeanstyle option pricing.91 30y 0.48 6m 1.65 0.94 0.17 0.21 0. tn))/A(t) is the forward par swap rate under consideration.91 15y 0.93 0. increases the rate at which the curvature of the smile decays with expiry.38 0.32 0.0 0% 20% 40% 60% 80% 100% –0.23 0.2 % mdl(t – 1m) mdl(t) ( ) dW2A (t ) (4) 0. Subnormal skews.50 0. ε.55 0.05 0.on a (discrete) basis of exponential functions.84 0. in many implementations of the SABR model the correlation between volatility and underlying rate are quite different for different expiries and tenors. In total.4 0.11 0. We note that the model’s volatility of variance parameter. The currency is the euro and the parameters were estimated from Totem consensus prices for the end of a particular month in 2004 1.45 0.27 0.52 0.05 0. and strikes from 5% to 95% in BlackScholes delta terms – all in all. We see that both models for the most part agree with the Totem consensus quotes within +/–0.0 0.20 10y 1.65 0.15 20y 1.18 15y 1. A.26 0.88 0. 912 swaptions were priced.32 0.90 20y 0. but our scope is to price general pathdependent instruments. t0) – P(t.49 0. Expiries range from six months to 20 years and tenors from one year to 30 years. the martingale measure with the annuity: A (t ) = ∑ δ i P (t . ti ) i =1 n δi = ti – ti – 1 is the day count fraction.84 0.29 1.13 1.26 5y 1. The first model had its m and ε grids fitted the month before whereas the second was fitted on the particular date.58 0. ε parameters fitted the month before the other had its parameters fitted this month.89 0. ε parameters is given in table A. which means that when the model is fitted to observed cap and swaption prices the implied volatility of volatility parameter most often turns out to be decreasing with the expiry of the underlying option.86 1.85 0.22 0.RISK. for a number of reasons. and all the parameters λ.27 1. The model is essentially a ‘shifted’ Heston (1993) model. so it allows for an analytic solution based on numerical inversion of the Fourier transform.14 0.14 0.6 –0. is related to lognormal volatility of volatility by the approximate relation: lognormal volatility of volatility ≈ ε / 2 This figure shows the deviations from Totem consensus quotes for euro swaptions in terms of implied Black volatility for two models.23 0. Expiries range from six months to 20 years. Nonzero correlation is technically quite difficult to handle in a full yield curve model and potentially timevarying correlation is of course even more complicated.0 where WA is a Brownian motion under annuity measure. This model is quite difficult to work with in the context of full yield curve models.95 0. corresponding to m < 0. Lipton (2002) and Lewis (2000) give representations of the option price that avoid the numerical instability of the representation in the original Heston (1993) paper. 912 swaptions. ε.97 5y 0.30 0.88 0.6 0. for two different models (4).31 0.62 0. β are constants.38 0.43 0. This model gives a good fit to observed cap and swaption prices with reasonably stable parameters across expiries and tenors.8 0.89 Stochastic volatility processes The most popular stochastic volatility model for caps and swaptions appears to be the SABR model by Hagan et al (2002) where the volatility is specified as a geometric Brownian motion that has some correlation with the underlying forward swap rate.38 0.05.83 0. figure 1 shows the deviations in terms of implied Black volatility from endofmonth Totem1 consensus quotes for euro swaptions for strikes ranging from 5–95% delta. As correlation between the swap rate and the volatility is assumed to be zero.12 0. All data is as of a particular end of month in 2004 In our experience.17 0. so in practice only the volatility level parameter λ needs to be updated on a regular basis. β.34 0. Increasing the speed of meanreversion. S(t) = (P(t.01 0. This in turn implies that a full dynamic version of the SABR model would have to exhibit even steeper decreasing forward volatility of volatility.NET ● SEPTEMBER 2005 RISK 105 .94 10y 0.86 0.75 0.13 0. the SABR model does not incorporate meanreversion in volatility. In terms of the implied BlackScholes volatility smile. The function κ → ηκ(t) can thus be viewed as the inverse Laplace transform of the forward rate volatility structure in the tenor dimension: t → σ(t.13 30y 1. Second. are possible with the – note of caution that S is restricted from above by 1 m m S(0) when m is negative. Instead we follow Andersen & Andreasen (2002) and use the following model as our basis for developing a full yield curve model with stochastic volatility: dS (t ) = λ z (t ) mS (t ) + (1 − m ) S (0) dW1A (t ) dz (t ) = β 1 − z (t ) dt + ε z (t ) dW1AdW2A =0 Note: this table reports best fit m and ε parameters to observed cap and swaption prices for β = 0.37 0. that is.58 0.
2×28. In this section. we do. Using Itô’s lemma and the fact that the swap rate S is a martingale under the annuity measure. τk. T ) ∈ » k . t + τ1 )′ ∈ » k × k 106 RISK SEPTEMBER 2005 ● WWW.4 0. .Cutting edge l Interest rates 2. τk. as in table A. T ) ρ (t . The currency is the euro 3 5 7 9 11 13 15 17 19 21 23 25 eps Here the matrix product RR′ is the instantaneous correlation matrix for the k forward rates. Model specification Andersen & Andreasen (2002) suggest a Libor market model with stochastic volatility.…. dZ (t ) dW (t ) = 0 (5) ( ) ( ( × λ (t .6 Legend? m 0. the model parameters that need to be set by calibration to swaption and cap prices are: ■ The forward rate volatility structure. SX = (∂S/∂X1.8 0. we let S be the 1×29 par swap rate for times between year zero and one. Though the implied skew parameter m increases over the period. Timeseries evolution of calibrated m. . however. ρ ( t . which is extended by Piterbarg (2003) to allow for a time. T ) ∈ » . t + τ k ) Equating diffusion terms of (6) and (7) yields: Γ (t ) η (t ) = z (t ) I M (t ) I F (t ) + I − I M (t ) I F (0) I λ (t ) R (t ) ⇓ −1 η (t ) = z (t )Γ (t ) I M (t ) I F (t ) + I − I M (t ) I F (0) I λ (t ) R (t ) (7) ( ) ( ) (7a) with: dz (t ) = β 1 − z (t ) dt + ε (t ) z (t )dZ (t ) . T ) = z (t . Swaption pricing For efficient calibration of the model. . T ) dz (t ) = β 1 − z 1 − z (t ) dt + z (t )ε (t ) dZ (t ) λ (t . at least in principle.0 0.and tenordependent local volatility skew parameter. The latter is typically estimated from the historical timeseries data of the yield curve. we get: dS (t ) = S (t )′ η (t ) dW A (t ) (9) X ( ) ( ) M (t ) = ( m (t . T ) f (t ) + 1 − m (t . m (t . though. can exactly fit the volatility level and slope for all expiries along k tenors.RISK. closedform pricing of caps and swaptions is essential.. . . the second the slope (skew) and the third the curvature (smile). t + τ1 ) . For the corresponding forward rates we have: dF (t ) = z (t ) I M (t ) I F (t ) + I − I M (t ) I F (0) I λ (t ) R (t ) dW (t ) + O ( dt ) where λ.. We note that m = 0 corresponds to a normal model whereas m = 1 corresponds to a lognormal model. The motivation for this is that if we consider implied parameters of the model (4). ∂S/∂Xk). In this case. On the other hand. dZdW = 0 ( ) (7b) m and ε for the 10year × 10year euro swaption over 2003–05. Approximate the stochastic differential equation (9) by the model: . τk. In practice. . 29×1 swaption smiles. . Our approximation goes in two steps: ■ A. … . T ) f (0.…... t + τ ) ∈ »k ×k Γ (t ) = ′ g (t .. t + τ n ) ′ ∈ » k 1 n k λt 1 n k ×k (6) R (t ) = ρ (t .…. T ) = 1 Z (t ) ∈ ». Fix k tenors τ1. the implied ε parameter appears to be fairly constant across both expiry as well as tenor.. . t + τ )) ∈ » F (t ) = f (t . T )′ dW (t ) + O ( dt ) )) The volatility specification (7) in combination with (3) defines our model.NET where we let subscripts denote partial derivatives. we typically see that the skew parameter m is fairly constant across expiry but it tends to decrease with tenor. ε over the period 2003–05.. λ for all times t and the tenors τ1. λ (t . our calibration will most often be on a best fit basis. t + τ1 )′ .. T ) m (t . whereas the curvature can only be fitted exactly for one tenor. . up to 29×1 par swap rate for times between year 28 and 29.2 This figure shows the evolution of the monthly calibrated 10year by 10year skew and smile parameters m. the implied volatility of variance ε is more or less constant over the period and both time series exhibit low volatility. . t + τ ) . λ and ρ are deterministic functions of time and maturity.. In terms of the implied BlackScholes volatility smiles for swaptions and caplets. ε for all times t. if we choose to fit the model to the strip of 1×29. ρ ( t . In most cases we choose constant κ1. For example. the first parameter controls the absolute level. we describe an accurate (near) closedform approximation. t + τ ) . 2×28 par swap rate for times between year one and two. ■ The forward rate skew structure.. the Piterbarg model can be formulated as: df (t . We see that the model. In this case we often specify the model a bit differently. t + τ ))′ ∈ » I ( ) = Diag (λ (t . . Under the separable volatility specification in (3). ε parameters 1. κk as well as a constant correlation structure RR′. ε is a deterministic function of time and β is a constant. ■ The forward volatility of volatility. that is. For the onefactor case. κk this derivative can be calculated in closed form by combining (7) with the bond price formula in (3). m are now scalar functions of time and S is a par swap rate referring to different swap periods over the time horizon. If we use continuously compounded rates rather than discrete rates as model primitives. ρ (t . m for all times t and the tenors τ1. namely: η (t ) = z (t ) m ( t ) S ( t ) + 1 − m ( t ) S ( 0 ) λ ( t ) (8) ( ) where m. often choose to go for an exact fit to a specific strip of swaptions or caplets. we have: dF (t ) = Γ (t ) η (t ) dW (t ) + O(dt ) ′ g (t . f (t .…. at least for expiries over one year. Given fixed meanreversion coefficients κ1. .2 0 1 –0. k = 1...
. As an example of this. We use approximation A and B to give us constant para~ ~ meters λ hj. Core areas include market and credit risk measurement and management.all remaining calculations are done in closed form. Articles that fail to meet the criteria are rejected at this stage. We use the methodology suggested by Piterbarg (2005ab). The calibration takes about five seconds of computer time. Approximation A essentially involves finding timedependent parame_ _ ters λ. the correlation structure for the forward rates RR′. His decision is final..( ) ( ) ( ( )) ( ) A dz (t ) = β (1 − z (t )) dt + ε (t ) z (t ) dW2 (t ) dS t = () A z t λ (t ) m t S t + 1 − m t S 0 dW1 t () () (10) where all parameters are timedependent. Depending on the feedback from referees. the pricing and hedging of derivatives and/or structured securities. This is probably due to the fact that we make no attempts to smooth our calibrated parameters in the time dimension. in addition to academia. . 0. Most often we calibrate the model in a sequence where only one of the weights γl. that is.. clarity of exposition is another yardstick for publication. However. This list is not an exhaustive one.com. brief communications address an extension or implementation issue arising from a fulllength article that. this technique is much faster and only marginally less accurate.. t + τi ) = mhi . . ~hj of the model (4) for these expiries ε (h) and tenors (j) of the calibration swaptions. Submissions should be sent to the technical team at technical@ incisivemedia.000 words.15. k. although Adobe PDFs are acceptable.30.. mi. We solve these by regression: m (t ) = ∑ S X (t ) ∂ X i =1 i k ∂ i 2 −1 z (t ) S X (t )′ η (t ) X = 0. does not deserve fulllength treatment. mhj. and for brief communications 1. We set: (κ1. Approximation B involves finding constant parameters so that the model (11) produces option prices that are close to those of (10) with parameters given by (13) and (15). ~hj for each swaption. exclusivity and relevance requirements. This is consistent with the findings in Piterbarg (2003). κ 4 ) = (0. 2 y.. The implied volatility smiles are parameterised by the parameters in table A. .. mhj. ■ B. We now calibrate the model by bootε strapping. 2. There does not appear to be a clear trend over time in any of the calibrated parameters. there is more noise in the calibrated forward skew parameters than in the Piterbarg (2003) case. that is: zλ 2 mS + (1 − m ) S (0) ≈ S X ′ η 2 2 (12) Equating levels in (12) at X(t) = Y(t) = 0 yields: λ (t ) = 2 1 S (0) 2 S X (t )′ Γ (t ) I F (0) I λ R −1 2 (13) X = 0.Y = 0 (14) for i = 1. We reserve the right to cut accepted articles to satisfy production considerations. It should be noted that this approximation can be slightly refined by evaluating (13) and (15) along levels of X. t + τi ) = λ hi . and the theoretical modelling and empirical observation of markets and portfolios. ε h }i =1. the method relies on a numerical solution of one Ricatti ordinary differential equation per swaption pricing .NET ● SEPTEMBER 2005 RISK 107 . WWW. γm. 0.30 y ) All quantities in (13) and (15) can be calculated in closed form using the zerocoupon bond price formula in (3). as in table A. ε (t ) = ε h for th – 1 < t ≤ th. We also welcome the submission of brief communications. γε is nonzero at the time. .Y = 0 k (15) sequentially for h = 1. τk of the model. risk management departments and trading desks of major financial institutions. Compared with a direct solution of (10) by numerical inversion of Fourier transform as suggested in Andersen & Andreasen (2002). Typically. γε are weights for balancing the different objectives against each other. . Here γl. Due to the form of η. Authors should allow four to eight weeks for the refereeing process. The resulting model parameters are shown in table B. Our referees are drawn from the research groups. Y corresponding to approximate expected levels of X. The exact details are quite complicated and are omitted and use a correlation matrix estimated for historical timeseries data of forward rate curves. the righthand_side of (14) is independent of z(t).k where all parameters are constant. We expect all articles and communications to contain references to previous literature.015.20) 2 2 λ ( t ) S ( 0 ) ∑ S X i (t ) X = 0.…. We assume that ~ ~ we have fitted parameters λ hj.RISK. Articles are then sent to one or more anonymous referees for peer review. and a set of tenors τ1.500 words. so (14) forms k linear equations in m. κk. Once received by the technical editor and his team. Many have already published articles in Risk.. The most important publication criteria are originality. m so that the diffusion in (9) is approximated by the diffusion in (10). Y under the annuity measure of the swaption under consideration.. are decreasing more sharply in tenor than the corresponding ‘term’ skew parameters shown in table A. We let the model (3) and (7) be parameterised by: λ (t . The preferred format is MS Word. and checked against the criteria above. Approximate the stochastic differential eqaution (10) by the timehomogeneous model: d S (t ) = z (t ) λ mS (t ) + 1 − m (t ) S (0) dW1A (t ) (11) A d z (t ) = β 1 − z (t ) dt + ε z (t )dW2 (t ) ( ) ( ) here for space considerations.1. submissions are logged. τ3 . we solve the optimisation problems: {λ hi . m (t . We see that the forward skew parameters.. The maximum recommended length for articles is 3. Given that Risk technical articles are shorter than those in dedicated academic journals. of expiries and a set of tenors {Tj} corresponding to the swaption smiles that we wish to calibrate the model to. with some allowance for charts and/or formulas. exclusivity and relevance – we attempt to strike a balance between these. Guidelines for the submission of technical articles Risk welcomes the submission of technical articles on topics relevant to our readership..Y = 0 i =1 (τ1. τ 4 ) = (6m. but the main point is that the technique is both very quick and accurate. κ 3 . These are also peerreviewed contributions to Risk but the process is less formal than for fulllength technical articles. Computationally. while satisfying our originality.. κ 2 . Calibration We start by fixing κ1.10 y. τ 2 . the technical editor makes a decision to reject or accept the submitted article. consider simultaneous calibration of a fourfactor model of the type specified in (3) and (7) to all the euro cap and swaption implied volatility smiles of 19 expiries ranging from six months to 20 years and eight tenors ranging from six months to 30 years. We further fix a time grid 0 = t0 < t1 < . . γm. Y = 0 Differentiating (12) with respect to Xi at X(t) = Y(t) = 0 yields: min γ λ ∑ λ hj − λ hj j ( ) 2 + γ m ∑ m hj − mhj j ( ) 2 + γ ε ∑ ε hj − ε hj j ( ) 2 (λ (t ) S (0) S 2 Xi (t ) X = Y = 0 ) 1 ∂ m (t ) = S X (t )′ η (t ) z (t ) ∂ X i 2 X = 0. mhi .
2464 0.82 0.2452 0.35 0.0658 0.97 0.2198 0. t + τi ) + c ( ( )) (16) where c is some constant. tenors range from six months to 30 years and strikes range from 5–95% BlackScholes delta – all in all 1.NET Strictly speaking.89 0.0805 0.1906 0.0416 0.6 –0.06 0. there is the error from the fact that a fourfactor model will not be able to exactly match the smiles of eight tenors. making the simulation error roughly of the order of 0.2798 0. Expiries range from six months to 20 years.0798 0.82 0.1207 0.0893 0.87 0.0631 m1(t) 0.2 –0.39 ε(t) 1. and comparing the resulting prices to those of the target model.1262 0.2120 0.1869 0.13 0.12 –0.1021 0. We call this “total calibration error” and the result of pricing up all the calibration swaptions by simulation is shown in Figure 108 RISK SEPTEMBER 2005 ● WWW.80 0.1335 0.0992 0.1148 0.1009 0.16 0.89 0.4 0.2266 0.2 % 0.1891 0.1664 0.26 0.50 –0.10 –1. however.85 0.77 1.1183 0.0873 0.65 –1.1325 0.0623 0.01 0.25 0. Pure calibration error 0. We see that the pure calibration error is within +/–0.01 0.1336 0.20 0.6 0.26 0. Moreover.20 0.40 1. CPU times for simulation in LMM and in separable HJM Maturity 5y 10y 15y 20y 25y 30y LMM 2.33 0.0602 0.47 0. the calibration only takes about five seconds of computer time.12 7. What actually counts.1147 λ2(t) 0.024 caplets and swaptions.89 –1.30 0.0907 0.25% in BlackScholes volatility terms in most of the range.0707 0.94 0.13 0. Parameters of a calibrated fourfactor model t 0.1278 0. min f (t .83 0.1492 0. We show this error by pricing swaptions and caps for all the calibration expiries and tenors by use of the approximation A and B.0784 0.20 0. We depict the difference between the yield curve model and the target when we price caps and swaptions by simulation.6 0.2242 0.92 0.05 –0.3078 0.4 0.88 –2.25 0. We see that the total calibration error is within +/0. In all.1268 0.2934 0. The strikes chosen correspond to 5–95% delta in BlackScholes terms.97 0. Jarrow & Morton (1992) and simply replace f(t.2954 0.22 0.14 0. expiries (six months to 20 years).0 0% –0.21 40.1413 0.53 6.8 –1.3064 0.1063 0.80 0.16 0.0872 0.5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 20 λ1(t) 0.24 0.0519 0.19 0.0734 0.19 26.87 0. 30y vanilla interest rate swaps with monthly reset in a fourfactor Libor market model and our fourfactor separable volatility structure HJM model 4.45 1.22 3.46 5.0410 0. f (0. 131.48 –1.3040 0.1219 0.2719 0.1678 0. straightforward Euler discretisation of the stochastic differential equation for z is going to exhibit very poor convergence as we decrease the time steps ∆t → 0.1410 0.33 4.94 0.75 0.0 This figure shows the pure calibration error in terms of implied BlackScholes volatility when calibrating a fourfactor model to the full euro cap and swaption market. We call this error ‘pure calibration error’ and it is shown in figure 3. and tenors (six months to 30 years).94 4.11 0.70 m3(t) –0.99 –1.04 1.08 0. We depict the difference between the yield curve model and the target when we price caps and swaptions under our approximations A and B.1490 0.86 0.1013 0. In summary. we price 1.01 –0.Cutting edge l Interest rates B.74 0. is of course what the error is when the model is simulated.0917 0. Instead. t + τi ) − c.10 0.22 0.01 –0. a fourfactor version of the model can simultaneously fit market prices of caps and swaptions for all strikes (5–95% delta). t + τi) in (7) with: f (t .4 –0.37 0.0529 0.01 –0. tenors from six months to 30 years and strikes from 5–95% in terms of BlackScholes delta – all in all 1.2693 0.36 –1.1705 0.4 –0.1671 0.10% in terms of implied BlackScholes volatility.05 0.1185 0.84 –1. t + τi ) .6 –0.1017 0.2 –0.1690 0.1278 0.84 0.42 –0.0768 λ3(t) 0. Due to the fact that the natural domain for the stochastic volatility factor z is {z ≥ 0}.1276 0.8 0.70 0.74 –1.44 –0.2038 0.0482 0.01 0.1254 0.024 caplets and swaptions 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Note: this table reports the resulting parameters when calibrating a fourfactor model to the euro swaption and cap data of table A 3.0 0.22 0.83 1.99 1. … .35 0.89 0.84 –1.90 1. To avoid this problem.40% in BlackScholes volatility terms in most of the range.13 HJM 1.07 –1.26 0. within a tolerance of 0.27 55.8 0.01 0.94 0.32 0.RISK.0 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% C.11 m4(t) –0.0635 0.81 0.024 caplets and swaptions Note: CPU times in seconds for simulation of 5y.1276 0.1246 0.8 This figure shows the total calibration error in terms of the BlackScholes volatility when calibrating a fourfactor model to the full euro market.56 –0.1162 0.97 2.0695 0.14 2. stochastic differential equations of the type defined by (3) and (7) can in some cases exhibit explosive behaviour. Expiries range from six months to 20 years.82 0.2 % 0.1236 0.02 1.2602 0. we follow Heath. we prefer to use the following (local) lognormal discretisation: .06 0.1554 λ4(t) 0.20 15.26 0. Monte Carlo simulation The error of such a calibration can be split in two. t + τi ) = max f (0.2541 0.28 –0.56 –0.1496 0.85 0.80 0.4% in implied Black volatility terms.03 m2(t) 0.1574 0. Total calibration error 1.84 0.95 0. First.1172 0.73 1.072 simulations were used.
accurate pricing can be obtained with monthly or quarterly time stepping. R Jarrow and A Morton. We use the standard threepoint discretisation for Dx and Dz. Squareroot processes such as (7b). This means that we get lower asymptotic accuracy than O(∆z2) but this does not seem to be a problem in practice. ( ) ( )( ) Finite difference solution For the onefactor model. pages 61–65 Mitchell A and D Griffiths. 2000 Turbocharging the Cheyette model Working paper. With typical parameter values.com WWW. ■ Jesper Andreasen is a principal in the fixed income quantitative research group at Bank of America in London. Y. Linear discretisation of the z axis according to the standard deviation of z at maturity leads to very few points in the interval [0. 2005b Stochastic volatility model with timedependent skew Forthcoming in Applied Mathematical Finance Ritchken P and L Sankarasubramaniam. 2005a Smiling hybrids Forthcoming in Risk Piterbarg V. 1993 On finite state Markovian representations of the term structure Working paper. ■ Accuracy of O(∆t2 + ∆x2 + ∆y4 + ∆zp). 1993 A closedform solution for options with stochastic volatility with applications to bond and currency options Review of Financial Studies 6. 1] relative to the number of points between one and the upper bound of z. 2003 A stochastic volatility forward Libor model with a term structure of volatility smiles Working paper. To illustrate this. In summary. say 10. 1992 Markov representation of the HeathJarrowMorton model Working paper. New York Piterbarg V. O(∆y4). 2002 Managing smile risk Wilmott Magazine. Attempting to solve this problem by transforming the state variable introduces infinite drift for the transformed variable at z = 0 and this is therefore not a recommendable route. A onefactor version of the model can be implemented with a finite difference solution and can thus be used as an alternative to the standard onefactor models for daytoday management of large portfolios of interest rate exotics. This gives higher accuracy in the y dimension. pages 327–344 Jamshidian F. pages 77–106 Heston S. The resulting computer times are reported in table C.ssrn. finite difference solution is an efficient alternative to Monte Carlo simulation. we have a scheme with the following properties: ■ Uniform von Neuman stability. 1993 Generalised Vasicek models of the term structure Applied Stochastic Models and Data Analysis 1.andreasen@bankofamerica. 1977 An equilibrium characterization of the term structure Journal of Financial Economics 5. The strength of the separable volatility structure relative to the general HJM or LMM specification is the speed in simulation of the model. and enables us to get away with relatively few y steps.NET ● SEPTEMBER 2005 RISK 109 . A Lesniewski and D Woodward. The associated pricing partial differential equation can be written as: ∂V 0= + Dx + Dy + Dz V ∂t r ∂ 1 ∂2 Dx = − + ( −κx + y ) + η2 2 3 ∂x 2 ∂x r ∂ Dy = − + η2 − 2κy 3 ∂y ( ) ∂ 1 ∂2 r Dz = − + β (1 − z ) + ε 2 z 2 3 ∂z 2 ∂z We use an alternating direction implicit scheme (see Mitchell & Griffiths.com Piterbarg V. 1980) that splits the solution over each time step into three steps: 1 1 2 1 1 ∆t − 2 Dx V t + 3 ∆t = ∆t + 2 Dx + Dy + Dz V (t + ∆t ) 1 1 1 1 2 1 ∆t − 2 Dy V t + 3 ∆t = ∆t V t + 3 ∆t − 2 DyV (t + ∆t ) 1 1 1 1 1 ∆t − 2 Dz V (t ) = ∆t V t + 3 ∆t − 2 DzV (t + ∆t ) (17) where V(t) is to be interpreted as a threedimensional tensor of values at time t. pages 131–170 Lewis A. 1992 Bond pricing and the term structure of interest rates: a new methodology for contingent claims valuation Econometrica 60. 1991 Bond and option evaluation in the Gaussian interest rate model Research in Finance 9. The disadvantage of the fivepoint discretisation is that the workload increases at a rate higher than the O(∆y–1) of a threepoint scheme but we find that is worth it in this particular case. July. D Kumar. Email: jesper. pages 163–168 Andreasen J. that is: z = 1 + e −β∆t z (t ) − 1 z (t + ∆t ) = ze − 1 v 2 + vN (0. with high volatility and low mean reversion and therefore high probability of hitting z = 0 can be tricky to solve numerically. pages 49–62 Cheyette O. pages 177–188 ( ) ε2 ε2 v 2 = ln 1 + z −2 1 − e −β∆t + z (t ) − 1 e −β∆t − e −2β∆t β 2β We combine this with standard Euler discretisation of X. k = 1. but for Dy we use a fivepoint discretisation for the first derivative. ■ Workload of O(∆t–1 × ∆x–1 × ∆y–q × ∆z–1). 2000 Option valuation under stochastic volatility Finance Press Lipton A. q > 1.1) 2 REFERENCES Andersen L and J Andreasen.RISK. a 30year Bermuda swaption is accurately priced on a grid of dimension 50 × 100 × 10 × 15 (t × x × y × z) steps and this takes about three seconds of computer time. we perform simulation of vanilla swaps with monthly rate reset in two models: an LMM with four factors and our separable model also with four factors. available from www. We see that in the LMM the computational time increases roughly with the square of the simulation horizon whereas it is linear for the separable model. pages 84–108 Heath D. p < 2. Conclusion We have presented a class of stochastic volatility yield curve models with quick and accurate calibration and significantly quicker Monte Carlo simulation than general HJM or Libor market models. Bank of America. University of Southern California Vasicek O. Table C and our experience indicate that one can obtain computational savings of up to a factor 10 for longerdated structures with the separable model relative to the LMM. 2002 Volatile volatilities Risk December. London. Barra Hagan P. In practice. 1980 The finite difference method in partial differential equations John Wiley. Instead we choose to discretise z according to zj O(j2). 2002 The vol smile problem Risk February._ where we choose z . Department of Finance. Gen Re Securities Babbs S. v so that the lognormal approximation matches the two first conditional moments of z(th + 1) given z(th).
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.